Attribution
Every token, tied to who caused it
Each model call is tied to the person, agent and run behind it. Projects, clients and cost centers are bound for a set period, so spend lands where it belongs.
Capability 03 · FinOps & Cost Control
Count every AI dollar. Cap it before the bill.
Cortx turns AI and cloud activity into money, puts each amount against the person, agent, project or cost center that caused it, warns before a budget runs out, and refuses further spend once it has.
Executive Cost OverviewTotal AI and cloud spend at a glance: burn rate, budget health and what's driving cost.
This month
$48,240Total spend · AI + cloud
$66,900Projected month-end
$1,040Unattributed, flagged
Model, cloud and GPU cost, in one place
Cortx prices the model calls it sees and pulls in the bills you already get, then puts every amount against the people and work that caused it.
Where spend comes from
Model providersOpenAI, Anthropic, Vertex, Bedrock, self-hosted
Cloud billingAWS Cost Explorer, Azure Cost Management, Google Cloud
On-prem clustersPrometheus and kube-state-metrics
GPUFrom contracts and cluster node hours
SaaS licensesMicrosoft 365, Google Workspace
FinOps & Cost Control
Where it lands
PeopleEvery model call tied to its user
Agents and runsSpend per agent, per run
Projects and clientsBound for a set period
Cost centersTeams, departments and tags
AlertsIn Slack and notifications, before the limit
Count a dollar once.
Stop it with the same count.
A dollar of AI spend should be counted once, in one ledger, against one price, and the budget that stops the next call should read that same ledger.
What every dollar is tied to
Personwho asked
Agent and runwhat acted
Projectbound for a period
Cost centerteam, department, tag
Modelpriced per token
Workspacewhere it ran
Four steps run on every model call. Each step hands the next what is on its arrow. Select a step to jump to it.
01 · Meter
Each model call's tokens are read from the provider's own response and priced. Cloud, on-prem and SaaS costs come in from the billing sources you connect.
02 · Attribute
Each call is tied to its user, agent and run. Project, client and cost-center tags come from bindings that hold for a set period.
03 · Budget
Budgets per account, user, agent, model, workspace, team, department or tag, per calendar period, with warn and critical thresholds.
04 · Stop
Before each model call, the agent runtime asks the budget gate first. Gateways add per-session and per-hour caps.
Five answers before every model call
The budget gate answers each call with one of five rungs, from carry on to stop everything.
- 1 Allow
- 2 Alert
- 3 Throttle
- 4 Block
- 5 Shut down
See where it goes.
Stop it where it should stop.
Budgets and alerts
Alerts before the limit, a stop at it
Set budgets by user, team, agent, model or tag. Alerts fire at your warn and critical thresholds, and once a budget is spent, the runtime refuses the next model call.
One view
Cloud, model and GPU cost together
Model spend sits next to AWS, Azure and Google Cloud billing, on-prem clusters and SaaS licenses. GPU cost comes from contracts and cluster node hours.
Cost control your CFO
can sign off on
How spend is counted, who it lands on, and what Cortx counts today.
Ask before spending
Every model call the agent runtime makes asks the budget gate first, and gets one of five answers: allow, alert, throttle, block or shut down.
Priced where it's metered
Tokens are read from the provider's own response and priced by the service that saw them, so the price on a call is the price it was charged at.
Buy through your cloud account
Xnode Cortx is on the AWS, Azure and GCP marketplaces, so it can go through your existing cloud account and procurement path.
What we count today
Model calls through the agent runtime and the model gateway are priced. Tool calls through the MCP gateway are counted and capped per agent and per user, but not priced. GPU cost is contracts and node hours, not GPU time per workload.
See your own spend, not ours.
We'll connect a billing source and walk through budgets, alerts and attribution on a scenario that looks like your firm.